Post
The classifier uses BLIP2 embeddings with training on labels which we manually created (yep). The ML team has the exact numbers but anecdotally it seems to have jumped from 40%ish accuracy to more like 90%. I'm surprised at how much more consistent it is.
3 reposts90 likes
Some things that may be of interest for reducing labeling requirements for specialized task, like topic analysis of social media posts 1. research.google/blog/achievi... (less training data) 2. developers.googleblog.com/en/introduci... (open weight model ready for fine-tuning on specialized tasks)